Comparative Analysis of CNN Methods for Periapical Radiograph Classification

Periapical radiographs are commonly used by dentists to diagnose dental problems and overall dental health conditions. The varying abilities of dentists to diagnose may be limited by their visual acuity and individual skills. To address this issue, there is a need for an application capable of computationally recognizing and classifying periapical radiographs. The commonly used computational method for image processing, specifically image recognition, is the Convolutional Neural Network (CNN) method. This study aims to create an application that can classify periapical radiographs and analyze the capabilities of the Convolutional Neural Network (CNN) method in this classification process. In general, periapical classification is divided into five types: Primary Endo with Secondary Perio, Primary Endodontic Lesion, Primary Perio with Secondary Endo, Primary Periodontal Lesion, and True Combined Lesions. The periapical radiograph classification process was tested using four CNN models: ResNet50v2, EfficientNetB1, MobileNet, and Shalow CNN. The evaluation of the CNN method utilized a confusion matrix-based technique to generate accuracy, precision, recall, F1-score and Weighted Average F1-score values. Based on the evaluation results, the highest accuracy value was achieved by EfficientNetB1 with 82%, followed by ResNet50v2 with 76%, MobileNet with 75%, and Shallow CNN with 71%.